{"id":"2bb69147-92ed-44eb-8434-ffa5c5250d38","arxiv_id":"2507.13995","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For each n from 8 through 2700 there is a locally area-minimizing three-chamber partition whose interface blows down to a singular cone, so the lens is not unique.","lead":"The authors construct stable three-chamber partitions of high-dimensional space whose interfaces minimize surface area yet end in singular cone shapes at infinity, not flat planes. This proves the standard lens cluster is not the only locally minimizing partition in every dimension from 8 to 2700.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.1 for n>8 rests entirely on the FLINT/Arb computation of Proposition 7.1; only n=8 has an independent by-hand check, so a code or identity bug would invalidate all n>8.","rationale":"The reader's weakest-assumption analysis identifies exactly the computational inequality in Proposition 7.1 as the bottleneck, and my reading confirms this. The analytic construction is detailed and internally coherent: the penalized minimization, concentration compactness, density estimates, and the strict-comparison contradiction in Section 6 are all plausible and no mathematical error surfaced there. However, the extension to n up to 2700 has no independent check except n = 8, and the by-hand n = 8 computation does not validate the odd-dimensional cases or the behavior of the margin near 2700. The closed forms (7.14)–(7.19) are intricate, and the computer code is the sole evidence for the strict inequality in higher dimensions. This is a genuine load-bearing concern because Theorem 1.2 is conditional on (1.5), and Theorem 1.1 is obtained by combining Theorem 1.2 with Proposition 7.1. The proposed concrete test directly targets this weakness by cross-checking with a different numerical route and by focusing on the largest, most sensitive dimension. For these reasons the reader's conditional verdict should stand.","tokens_in":36544,"tokens_out":36971,"duration_ms":357451,"concrete_test":"Run the supplied FLINT/Arb program at n = 2700 with increased precision (e.g., 256-bit mantissa) and independently recompute M(k,l) and Λplane(n) by direct interval integration of (7.12)–(7.13) instead of the closed-form special-function expressions; require the margin Λplane(n) − M(k,l) to be positive with an interval lower bound above 10^-30. If both implementations agree with a positive margin, the numerical hypothesis is confirmed; if the margin is not resolvable at that precision or the two implementations disagree, Theorem 1.1 is not established for n > 8.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The load-bearing step is Proposition 7.1: the strict inequality ΛLawson(n) < Λplane(n) for every n ∈ {8,...,2700}. The analytic proof of Theorem 1.2 is conditional on exactly this comparison, and the only independent by-hand verification is n = 8 (Section 7.4). For all other dimensions the inequality depends on interval-arithmetic evaluation of the closed forms (7.1) and (7.14)–(7.19) using the FLINT/Arb C code. An unnoticed bug in that code, or an error in one of the hypergeometric or Appell-function identities, would break Theorem 1.1 for every n > 8. The margin Λplane(n) − M(k,l) is reported to decrease with n up to 2700, so the largest dimensions are the most sensitive. The paper's own remark after Proposition 7.1 states that the FLINT code cannot estimate Λplane(n) reliably past roughly 2800, underscoring that the verification is near the limits of the numerical method. No part of the analytic argument in Sections 3–6 compensates for this reliance on a single computational pipeline.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper constructs, for each n in {8,...,2700}, a locally minimizing (1,2)-cluster in R^n that is not the standard lens cluster, answering a question raised in earlier work by Bronsard and Novack. The construction is variational: for each large R, a penalized perimeter problem is solved in a ball with boundary data given by a singular area-minimizing cone K, and a compactness/concentration argument yields a limiting locally minimizing cluster. The main conditional theorem (Theorem 1.2) states that if the strict inequality Λ(∂K) < Λplane(n) holds, then the resulting cluster has a blowdown whose interface is a singular area-minimizing cone. The paper then verifies this strict inequality for the Lawson cones, using a computer-assisted interval-arithmetic computation (Proposition 7.1), and gives a complete by-hand verification for n = 8. The paper also contains a related existence result for a capillarity problem inside cones.","tokens_in":36775,"tokens_out":5034,"duration_ms":57633,"significance":"If the proofs are correct, this is a significant advance: it establishes non-uniqueness of the standard lens cluster in all dimensions from 8 to 2700 and shows that singular area-minimizing cones can appear as blowdowns of local perimeter minimizers in the (1,2)-cluster problem. The analytic framework in Sections 3–6 is coherent: the penalized minimization, concentration compactness, density estimates, monotonicity formula, and the energy comparison argument are all laid out in detail. The by-hand computation of Λplane(8) and M(3,3) in Section 7.4 is a useful, independently checkable special case. The paper also contains a number of clearly stated limitations (e.g., the blowdown K∞ is not identified with the original cone K, and the numerical verification stops at n = 2700), which is commendable for its honesty.","major_comments":[{"comment":"Proposition 7.1 is the sole computational foundation for Theorem 1.1 for n > 8, yet the manuscript does not report the actual certified interval enclosures or the precision parameters used in the FLINT/Arb computation. In particular, the margin Λplane(n) − M(k,l) decreases with n, and the paper states that the code cannot estimate Λplane(n) past roughly 2800, so the verification near n = 2700 is especially delicate. The reader needs to see the interval widths for the largest n and a reproducible script that outputs the enclosures. Please include a table of rigorous upper and lower bounds for Λplane(n) and M(k,l) for a sample of n (or for all n in the range), along with the precision and the version of the Arb library used.","section":"§7.3, Proposition 7.1"},{"comment":"The closed forms (7.14)–(7.19) rely on the Euler and Picard integral representations, whose validity requires the parameter conditions stated in Appendix B (e.g., max{|λ/(ρ−d)|, |λ/(ρ+d)|} < 1). The text asserts these inequalities hold 'by construction', but no quantitative verification is provided for the full range n ∈ {8,...,2700}. Since the formulas are used to define M(k,l), a failure of these conditions for some n would invalidate Proposition 7.1. Please add a short argument or a numerical check showing that these inequalities hold for every k,l considered in Proposition 7.1.","section":"§7.2, Appendix B"},{"comment":"Equation (6.6) contains a typographical error: 'Hn−1(∂K ∩ ¯X (1)(1))' should presumably read 'H^{n-1}(∂K ∩ ¯X(1))'. More substantively, the gluing construction that leads from (6.6) to the contradiction with minimality is only described in words ('a repetition of the same gluing argument giving us (6.4) above'). Since this is the step where the strict inequality Λ(∂K) < Λplane(n) is actually used, the gluing should be formulated precisely, including the control of the error term o_k(1). Without this, the proof of Theorem 1.2 is not fully verifiable.","section":"§6, Step 2"}],"minor_comments":[{"comment":"The definition of gR is written with two cases that overlap: 't − √R / √R for t ≥ √R' and '0 for t ∈ [0, R)'. The second case should be '0 for t ∈ [0, √R)' (or the first case should be 't ≥ R'), since the later estimates such as (3.8) rely on gR being positive for t ≥ √R.","section":"§3, Eq. (3.1)"},{"comment":"The phrase 'P(Xlens(1)) − ω_{n−1} ρ_n^{n−1} / n' is not self-explanatory; the notation is only clarified later in (7.4). Consider adding a sentence explaining that Λplane(n) is the renormalized energy after subtracting the area of the interface disk in the planar part of the lens.","section":"§1.1, Eq. (1.3)"},{"comment":"Table 2 lists exact values of Λplane(n) for n = 8,...,16, but the derivation is only sketched via the recurrence in (7.23). A brief note on how the closed forms were simplified (e.g., using the contiguous relation and evaluating at z = 1/4) would help the reader reproduce these entries.","section":"§7.4, Table 2"},{"comment":"The reference to [29, Chaper 12] contains a typo: 'Chaper' should be 'Chapter'.","section":"§2, Notation"}],"recommendation":"major_revision","confidential_remarks":"The main concern is the heavy reliance on the FLINT/Arb computation in Proposition 7.1. The paper does include the C source code as supplementary material and gives a by-hand check for n = 8, which is good. However, given that the code was written with the assistance of ChatGPT (as acknowledged) and that the margin shrinks with n, I would recommend that the editor ask for an independent verification of the numerical inequality for a few large n, or at least that the authors provide a detailed output table with certified enclosures. The analytic core of the paper appears sound, and the n=8 case is a valuable nontrivial result by itself."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a good paper that settles a real open problem. It constructs, for n = 8,...,2700, locally minimizing (1,2)-clusters that are not standard lenses, with singular area-minimizing cones as blowdowns. The main analytic engine is a penalized minimization on balls with boundary data from a singular cone, followed by a concentration-compactness argument. The new trick is that they do not try to rule out loss of compactness along the singular interface; instead they show that such a loss produces exactly the kind of minimizer they want. The argument is coherent and detailed, and the density estimates, almost-monotonicity, and contradiction argument are all laid out properly.\n\nThe by-hand computation for n = 8 is a real virtue: they give exact closed forms for Λplane(8) and M(3,3) and check the strict inequality without any computer. That gives independent support. The FLINT/Arb code for n up to 2700 is shipped in the supplementary files, and the paper is transparent about the interval arithmetic and its limit at around 2800.\n\nThe soft spot is exactly what the reader flagged: for every n > 8, Theorem 1.1 rests on the computer-assisted verification of Proposition 7.1. If there is a bug in the C code or in one of the hypergeometric/Appell identities, the theorem breaks for all those dimensions at once. I do not see any sign of such a bug, and the n = 8 check plus the Mathematica cross-checks up to n = 36 temper the worry, but it is still a single pipeline. A referee should run the code or at least independently evaluate a few dimensions. The other caveat is that Theorem 1.2 does not identify the blowdown cone; it just guarantees some singular cone. That is a structural limitation the authors openly discuss, not a flaw in the proof. Section 8 on capillarity is proof-by-summary, but it is a side result.\n\nI would rate this as a solid paper. The central argument is convincing, the computational part is honest and reproducible in principle, and the result is significant within the cluster and minimal surface community. I would send it to a serious referee, especially one willing to check the numerics.","headline":"New construction of locally minimizing non-standard lens clusters in dimensions 8–2700, with a clean analytic proof and honest computer-assisted verification; the n>8 range hangs on the FLINT code, but the paper deserves peer review.","tokens_in":37306,"tokens_out":2311,"would_cite":true,"duration_ms":26040,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49Q20","49Q05","53A10"],"pacs":[],"model":"deepseek-v4-flash","headline":"In every dimension from 8 to 2700 there exists a locally minimizing three-chamber partition of space that is not the standard lens cluster and whose interface at infinity is a singular cone.","keywords":["locally minimizing clusters","(1,2)-cluster","standard lens cluster","singular area-minimizing cones","Lawson cones","non-uniqueness","computer-assisted proof","interval arithmetic"],"falsifier":"Run an independent interval-arithmetic evaluation of formulas (7.1) and (7.14)--(7.19) for each $n = 9, \\dots, 2700$; if for any such $n$ the computed competitor energy $M(k,l)$ is not strictly below $\\Lambda_{\\mathrm{plane}}(n)$, then Theorem 1.1 fails in that dimension. The $n=8$ case is separately settled by the paper's by-hand computation, so it would survive such a failure.","tokens_in":36324,"feed_emoji":"📐","tokens_out":7767,"duration_ms":82155,"temperature":0.7,"pith_summary":"This paper sets out to show that the standard lens cluster—the symmetric three-chamber partition of $\\mathbb{R}^n$ into a bounded droplet and two infinite chambers that locally minimizes interfacial area—is not the only locally minimizing cluster once the dimension reaches 8. The authors build rival clusters by forcing the outer interface to follow a singular area-minimizing cone at large scales and then letting the scale go to infinity. They prove that in every dimension from 8 to 2700 such a rival exists and that its blowdown at infinity is a singular cone. If the construction is right, the classification of locally minimizing three-chamber clusters in low dimensions is genuinely sharp, and singular cones appear as asymptotic shapes rather than only as defects.","feed_headline":"Standard lens clusters fail to be unique in dimensions 8–2700","feed_subtitle":"Locally minimizing three-chamber partitions can instead blow down to singular cones built from Lawson cones.","key_machinery":"The machinery is a renormalized energy comparison between the lens constant $\\Lambda_{\\mathrm{plane}}(n)$, the asymptotic cost of placing a lens-shaped droplet on a flat interface, and the cone constant $\\Lambda(\\partial K)$, the infimum cost of placing a unit-volume droplet near an interface modeled on a singular cone $\\partial K$. The proof runs a sequence of penalized energy minimizers in balls of radius $R$ with boundary data $K$ outside $B_{3R}$; the penalization confines the droplet while its Lipschitz constant $1/\\sqrt{R}$ dies in the limit, so any limiting concentration minimizes the unpenalized perimeter. A partial concentration-compactness argument shows that if all mass escaped as lens-shaped droplets the total energy would be at least $\\Lambda_{\\mathrm{plane}}(n) + P(K; B_{4R})$, contradicting the strict inequality $\\Lambda(\\partial K) < \\Lambda_{\\mathrm{plane}}(n)$; therefore some concentration survives with singular blowdown. The numeric input is Proposition 7.1, which evaluates the closed forms (7.1) and (7.14)--(7.19) in interval arithmetic.","core_discovery":"The central claim is Theorem 1.1: for each $n \\in \\{8,\\dots,2700\\}$ there is a locally minimizing $(1,2)$-cluster $X$ that is not the standard lens. The mechanism is a conditional statement, Theorem 1.2: whenever $K$ is a singular perimeter-minimizing cone in $\\mathbb{R}^n$ and the renormalized energy comparison $\\Lambda(\\partial K) < \\Lambda_{\\mathrm{plane}}(n)$ holds, the penalized minimization procedure yields a locally minimizing cluster whose blowdown interface is a singular area-minimizing cone. Theorem 1.3 verifies the comparison in the stated dimensions using the Lawson cones $C_{k,l}$: with $k = n/2-1$ for even $n$ and $k=(n-3)/2$, $l=k+1$ for odd $n$, the paper computes the competitor energy $M(k,l)$ and proves $M(k,l) \\le \\Lambda_{\\mathrm{Lawson}}(n) < \\Lambda_{\\mathrm{plane}}(n)$ by rigorous interval arithmetic, with a full by-hand verification for $n=8$. Hence the standard lens cluster is not the unique local minimizer in these dimensions.","pith_inferences":["The numerical gap $\\Lambda_{\\mathrm{plane}}(n) - M(k,l)$ appears to shrink as $n$ grows, so a plausible route to all dimensions $n \\ge 8$ is to prove monotonicity of $\\Lambda_{\\mathrm{plane}}$ and then verify the inequality by asymptotic expansion rather than computation.","If the strict comparison persists in every dimension, the same construction should give non-uniqueness of the standard lens cluster in all $n \\ge 8$, although the specific competitor used here may cease to work at very large $n$.","The blowdown of the constructed cluster is not shown to be the same cone $K$ used as boundary data; identifying $K_\\infty$ would require uniqueness of tangent cones to minimal surfaces, which the paper notes is open.","Because the energies are continuous in the surface-tension weights, the same construction should produce non-standard minimizers for weighted cluster energies near equal weights, a direction the paper only sketches."],"forward_implications":["The classification in dimensions $n \\le 7$ cannot extend to $n \\ge 8$: for every $n = 8, \\dots, 2700$ there are at least two non-homothetic locally minimizing $(1,2)$-clusters.","The singular area-minimizing cones that were already known to exist become genuine asymptotic profiles of local minimizers, not merely obstructions to regularity.","In even dimensions the energy comparison can in principle be checked by hand; the paper carries this out explicitly for $n=8$, giving an independent verification of the first new dimension.","The same penalized-compactness scheme supplies minimizers for a capillarity problem inside singular cones under an analogous strict inequality, as described in Section 8."],"supporting_citations":[{"why":"Establishes that non-planar area-minimizing hypercones exist in $\\mathbb{R}^8$, the seed construction used throughout Theorem 1.2.","marker":"[8]"},{"why":"Gives the $n \\le 7$ classification and the planar-growth rigidity for the standard lens that the new clusters must beat.","marker":"[10]"},{"why":"Proves that the family of Lawson cones $C_{k,l}$ is area-minimizing in the needed dimensions, supplying the explicit cones for Theorem 1.3.","marker":"[22]"},{"why":"Supplies arbitrary-precision interval arithmetic used for rigorous error estimates in Proposition 7.1.","marker":"[21]"},{"why":"Provides the FLINT library that evaluates the closed-form special-function expressions to certify $M(k,l) < \\Lambda_{\\mathrm{plane}}(n)$.","marker":"[47]"},{"why":"Allard's regularity theorem sets the density threshold that separates planar from singular blowdowns in Case 1 of Theorem 1.2.","marker":"[4]"},{"why":"Provides the locally isoperimetric partition framework used in constructing admissible competitors and passing to limits.","marker":"[36]"}],"fun_headline_variants":["Lens cluster uniqueness fails in 8 to 2700 dimensions","Singular cones break lens cluster uniqueness up to 2700D","Non-unique minimizing clusters via singular cones in 8–2700","Standard lens not the only minimizer in high dimensions","Local minimizers: lens cluster non-unique from dimension 8"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the interval-arithmetic computations certifying $\\Lambda_{\\mathrm{Lawson}}(n) < \\Lambda_{\\mathrm{plane}}(n)$ for every $n = 8, \\dots, 2700$ are correct; only the $n=8$ case has an independent by-hand verification, so a bug in the code or in the special-function identities would undo the proof for every dimension above 8.","fun_headline_variants_meta":{"raw":{"variants":["Lens cluster uniqueness fails in 8 to 2700 dimensions","Singular cones break lens cluster uniqueness up to 2700D","Non-unique minimizing clusters via singular cones in 8–2700","Standard lens not the only minimizer in high dimensions","Local minimizers: lens cluster non-unique from dimension 8"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000793,"raw_usage":{"total_tokens":3448,"prompt_tokens":854,"completion_tokens":2594,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":470,"completion_tokens_details":{"reasoning_tokens":2505}},"tokens_in":470,"tokens_out":2594,"duration_ms":21017,"temperature":1.0,"reasoning_tokens":2505,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T16:11:14.210895+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run an independent interval-arithmetic evaluation of formulas (7.1) and (7.14)--(7.19) for each $n = 9, \\dots, 2700$; if for any such $n$ the computed competitor energy $M(k,l)$ is not strictly below $\\Lambda_{\\mathrm{plane}}(n)$, then Theorem 1.1 fails in that dimension. The $n=8$ case is separately settled by the paper's by-hand computation, so it would survive such a failure.","supporting_citations":[{"cited_title":"Bombieri, E","cited_arxiv_id":null,"evidence_quote":"Establishes that non-planar area-minimizing hypercones exist in $\\mathbb{R}^8$, the seed construction used throughout Theorem 1.2."},{"cited_title":"Bronsard and M","cited_arxiv_id":null,"evidence_quote":"Gives the $n \\le 7$ classification and the planar-growth rigidity for the standard lens that the new clusters must beat."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proves that the family of Lawson cones $C_{k,l}$ is area-minimizing in the needed dimensions, supplying the explicit cones for Theorem 1.3."},{"cited_title":"Johansson, Arb: efficient arbitrary-precision midpoint-radius interval arithmetic , IEEE Trans","cited_arxiv_id":null,"evidence_quote":"Supplies arbitrary-precision interval arithmetic used for rigorous error estimates in Proposition 7.1."},{"cited_title":"FLINT team, FLINT: Fast Library for Number Theory , 2025","cited_arxiv_id":null,"evidence_quote":"Provides the FLINT library that evaluates the closed-form special-function expressions to certify $M(k,l) < \\Lambda_{\\mathrm{plane}}(n)$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Allard's regularity theorem sets the density threshold that separates planar from singular blowdowns in Case 1 of Theorem 1.2."},{"cited_title":"Novaga, E","cited_arxiv_id":null,"evidence_quote":"Provides the locally isoperimetric partition framework used in constructing admissible competitors and passing to limits."}],"review_version":1}